Lennon Crow’s Personal Website

Personal Projects

Portfolio Website

  • Designed and deployed personal portfolio through two iterations, demonstrating continuous improvement and evolving technical architecture.
  • Iteration 1: Built initial deployment on Raspberry Pi running headless Ubuntu server using Flask, Gunicorn, and Apache with secure HTTPS/SSL certificate, implementing DDoS protection and IP blocking through firewall rules, managed via SSH.

GitHub: View Flask-Based Personal Website on GitHub

  • Iteration 2: Migrated to virtualized Ubuntu Server environment using VirtualBox, deploying WordPress for improved content management, customization, and security.
  • Integrated custom CSS/HTML/JavaScript for video playback functionality and implemented reCAPTCHA API to protect contact forms from bots.
  • Configured enterprise-level network isolation using managed switch, VLANs, and network bridges to isolate web server VM, implementing VLAN segmentation that restricts internal access while maintaining public accessibility.

DIY Home Thermostat with Home Assistant

  • Designed and built custom thermostat hardware directly integrated with HVAC system.
  • Developed distributed temperature monitoring system using multiple ESP32 sensors and MQTT protocol to stream real-time data for whole-house average calculations.
  • Programmed automation logic in Home Assistant to control HVAC based on aggregate temperature data.
  • Implemented state-based fail-safe rules preventing simultaneous heating/cooling and compressor damage.

MotionEyeOS Security System

  • Designed and implemented a home security system using Raspberry Pi and MotionEyeOS.
  • Configured a camera-based motion detection system and set up a dedicated Ubuntu Linux server for video storage using SSH.
  • Demonstrated proficiency in network configuration, system administration, and open-source software integration.

Academic Projects

Accessible Indoor Navigation System, DFW Airport

  • Led 4-person team through three development iterations, coordinating task assignments and presenting the final demo, earning the course’s Best Implementation award.
  • Designed the pipeline converting a DFW Terminal D floor plan into a routable database: nodes placed over the plan image, georeferenced from three known-coordinate control points, connected with 8-directional edges, and exported as generated SQL, yielding a 1,351-node / 5,944-edge graph with 129 mapped amenities.
  • Built calibration tooling overlaying the stored graph on Google Maps so node placement drift was visible and correctable rather than guessed at.
  • Delivered accessibility-aware routing over PostgreSQL with the pgRouting extension on AWS RDS, supporting wheelchair-accessible and gender-specific restroom matching behind a 20+ endpoint Flask REST API.
  • Enforced secure/public zone separation so generated routes never cross the TSA boundary, a constraint that invalidates naive shortest-path results inside a terminal.
  • Developed Kotlin Android application using Google Maps SDK with turn-by-turn guidance, off-path recalculation, and rerouting when amenities close or reach capacity.
  • Owned requirements and verification: 36 functional requirements traced across 21 use cases, three IEEE-format test plans, and 87 automated tests organized by use case so each maps back to an agreed requirement.
  • Served as class representative to faculty, advocating cohort-wide on project scope and scheduling, and produced the SRS, project plan, and architecture, component, and deployment diagrams.

GitHub: View Accessible Indoor Navigation System on GitHub

Interactive Smart Home Dashboard

  • Integrated React/TypeScript frontend with a Flask backend, sending device state changes to the API and displaying the state returned by the backend across seven device categories.
  • Serialized concurrent writes to the shared state store behind a locking queue, eliminating corrupted reads when multiple device changes landed at the same time.
  • Built the event-logging subsystem, persisting per-device state changes with attributed reasons and surfacing a scrollable 50-event activity history in the dashboard.
  • Built the notification system that served as the manipulated variable in a 5-participant usability study, implementing all three conditions tested: activity log only in iteration one, Discord webhook alerts in iteration two, and browser push notifications in iteration three.
  • Deployed all three prototype iterations simultaneously, allowing participants to be tested across every version.
  • Co-designed Likert questionnaire measuring user trust, transparency, and usability, and moderated 2 of 5 participant sessions.

GitHub: View Interactive Smart Home Dashboard on GitHub

DistilALBERT: Compressing BERT with Knowledge Distillation

  • Led 3-person team compressing BERT-base (109.5M parameters) into a smaller ALBERT-based model using knowledge distillation.
  • Implemented four-part training objective combining masked language modeling, KL divergence against a frozen BERT teacher, cosine hidden-state alignment, and sentence-order prediction.
  • Pre-trained the model on 500,000 English Wikipedia documents using a single NVIDIA A30 GPU, reducing masked language model perplexity from 23,710 to 977.
  • Produced a 45.3 MB checkpoint at 11.8M parameters, a 9.3x reduction from BERT-base, scoring 0.807 accuracy on GLUE SST-2.
  • Benchmarked results against BERT-base, DistilBERT, and ALBERT-base-v2 on SST-2 and MRPC classification tasks.

Dallas County Property Valuation ML Predictor

  • Built end-to-end machine learning pipeline predicting residential property values using 3.1M+ records from Dallas Central Appraisal District public data (2021–2025), achieving R² of 0.87.
  • Developed modular six-stage Python ETL pipeline automating web scraping, ZIP extraction, multi-table SQL-style joins, categorical encoding, and data validation.
  • Integrated three government data sources (DCAD, Dallas Open Data, Texas Education Agency) using composite keys and pandas merge operations with configurable join strategies.
  • Performed feature selection using Pearson correlation for numeric variables and eta-squared for categorical variables, identifying ZIP code (η² ≈ 0.31) and school district (η² ≈ 0.22) as top predictors.
  • Trained and benchmarked five regression models (Linear Regression, Decision Tree, Random Forest, XGBoost, LightGBM), leveraging GPU acceleration to reduce gradient boosting training time from 10+ minutes to under 15 seconds.

Open-Source Secure Multi-Modal AI Platform

  • Led 6-person engineering team through complete software development lifecycle.
  • Provisioned and secured AWS EC2 GPU instances to host 7B to 14B parameter models.
  • Implemented and tested open-source solutions such as Ollama and Open WebUI, ensuring alignment with sponsor expectations.
  • Developed end-to-end setup script automating GPU detection, dependency installation, security patches, LUKS-encrypted Docker storage, and service configuration.
  • Held weekly stakeholder meetings to present sprint progress and gather feedback.

KWIC Web Search Engine

  • Developed Python-based search engine using Key Word in Context (KWIC) indexing with Flask framework.
  • Utilized AWS RDS with MySQL backend for scalable storage, indexing over 100,000 entries.
  • Implemented multi-threading for concurrent request handling.
  • Integrated advanced features including autocomplete suggestions and dynamic result ordering based on access frequency.
  • Collaborated with team using GitHub for version control and delivered stakeholder presentations.

GitHub: View KWIC Web Search Engine on GitHub

Requirements for an Indoor Navigation App for the Visually Impaired

  • Conducted stakeholder analysis, identifying primary and secondary stakeholders, including visually impaired users, caregivers, and developers.
  • Defined and validated functional and non-functional requirements, such as real-time audio feedback, hazard detection, system usability, customization options, and route optimization for user comfort and safety.

Database and Webpage Project

  • Developed a user-friendly website using HTML, Bootstrap (frontend) and Python, Flask (backend).
  • Employed SQLAlchemy to handle database queries and data storage efficiently.
  • Successfully delivered a fully functional system showcasing strong web development, database design, and application integration skills.

GitHub: View Database Project on GitHub

Project Management: Mock Google Budget Integration

  • Collaborated on managing the project of a mock Google Budget application, managing hypothetical resources, including personnel and salary allocations.
  • Utilized Microsoft Project for scheduling, budgeting, and risk management.
  • Implemented project layering techniques to streamline the development timeline and optimize resource utilization.

Multithreading Simulation in a Doctor’s Office

  • Created a multithreading simulation where threads represented different entities in a doctor’s office (patients, receptionist, nurses, and doctors) interacting with each other.
  • Debugged complex threading issues and ensured smooth interaction between multiple threads representing real-world processes.

GitHub: View Multithreading Simulation on GitHub

CPU and Memory Interaction Simulation

  • Developed a simulation of CPU and memory interaction using pipes and forks to create two processes that fetched and processed commands, simulating a basic CPU architecture.
  • Debugged issues related to pipes messaging and wait systems; adapted the project to Linux due to the lack of support for forks and pipes in Windows.

GitHub: View CPU and Memory Simulation on GitHub